XBDM hack is slow

[JayFoxRox]

Current code (eb69c7fba9cdd1bb1e81eec60f4fed4394a42c67):

def xbdm_hack(address, operation, data=0):
  SetMem(hack_bank, struct.pack("<III", address, operation, data))
  xbdm_command("resume thread=0x" + format(hack_bank, 'X'))
  return GetMem(hack_bank + 8, 4)

(Also see xbdm-hack.md for more information)

Here is what the hack does for reads and writes:

Reads

def xbdm_read_8(address):
  return xbdm_hack(address, 1)

1. xbdm_hack will call SetMem, which is the first communication with Xbox, to setup the next step
2. xbdm_hack will call xbdm_command which is another communiation with Xbox
3. xbdm_hack will readback the result which is another communication with Xbox

That is 3 back-and-forth transfers for a single read.

Writes

def xbdm_write_8(address, data):
  xbdm_hack(address, 4, int.from_bytes(data, byteorder='little', signed=False))

1. xbdm_hack will call SetMem, which is the first communication with Xbox, to setup the next step
2. xbdm_hack will call xbdm_command which is another communiation with Xbox
3. xbdm_hack will readback the result which is another communication with Xbox, even if it isn't returned from xbdm_write.

That is 3 back-and-forth transfers for a single read, with one of them being completly useless.
There's various ways to optimize it.

Calls

def xbdm_call(address, stack):
  assert(len(stack) < 64)
  SetMem(hack_bank + 12, stack)
  return xbdm_hack(address, 7, len(stack))

1. xbdm_call will call SetMem, which is the first communication with Xbox
2. xbdm_hack will call SetMem, which is the more communication with Xbox, to setup the next step
3. xbdm_hack will call xbdm_command which is another communiation with Xbox
4. xbdm_hack will readback the result which is another communication with Xbox

---

Ideally we'd pass all optional input data in the xbdm_hack request, and return all optional output data in the response.
Maybe we should also allow packing of commands, simply to avoid XBDM command processing overhead.